max.eth

Luca’s mission is ultimately straightforward: show the world the greatness of @pudgypenguins. Not as a speculative asset, and not as an artifact of one particular market cycle, but as a globally recognisable universe with characters, products, stories, and a community people genuinely want to belong to. That is the deeper test for crypto-native brands: can they turn attention into affinity, and affinity into something durable? With Pudgy World and Overpass on the horizon, Pudgy Penguins is continuing to build toward that vision.

For Luca, the ambition has never been to make Pudgy Penguins successful only within crypto.🐧 The larger objective is to build a mass-market brand, one that can reach people who have never traded an NFT, used a wallet, or followed the culture that produced the project. That means meeting people where they already are: through products, retail, characters, entertainment, and experiences that do not require technical literacy to understand.

On this episode of @Web3Rehashed, we speak with @LucaNetz, CEO of @pudgypenguins, about one of crypto’s most difficult questions: how does an internet-native project become a brand that can survive beyond the cycle that created it? Pudgy Penguins launched on Ethereum in July 2021. In April 2022, Luca acquired the project for 750 ETH, then set out to transform it from a well-known NFT collection into a broader consumer-facing company. Today, Pudgy products sit in Walmart stores across the United States, an outcome that would have sounded improbable for a PFP project only a few years ago. This conversation explores the decisions, risks, and long-term thinking behind that transition.

Two days ago, Anthropic launched Claude Fable 5 and Claude Mythos 5. On the surface, this looks like another frontier model release. But I think the more important story is not just capability. It is access. Fable 5 is the public Mythos-class model. Mythos 5 is the same underlying model, but with some safeguards lifted for trusted cyber defenders, infrastructure providers, and eventually select biology researchers. In its early stages, Anthropic is demonstrating what the next phase of frontier AI deployment may look like👇 ~~ Analysis by @punkbennet ~~ I, like many others, have become slightly numb to model launches. Every few months, a new model arrives with better coding, better reasoning, better long-context performance, better benchmark charts, better agentic workflows, better everything. At some point, the launch cycle starts to blur into one long benchmark war. But the Fable / Mythos release feels different. Not because Anthropic is claiming another step forward in intelligence, though it is. Not because the model seems strong at long-horizon coding, scientific reasoning, vision, finance, and complex knowledge work, though that matters too. It feels different because Anthropic is openly splitting capability into two layers: a public version and a trusted-access version. That is the real story. Fable 5 is described as a Mythos-class model made safe for general use. According to Anthropic, it exceeds any model they have previously made generally available and is especially strong on long, complex tasks. The model is available to general users, but it ships with classifiers that detect certain categories of high-risk use. When those classifiers trigger, the request does not get handled by Fable 5. It falls back to Claude Opus 4.8. The covered areas are cybersecurity, biology and chemistry, and distillation. In plain language: domains where the model’s raw capability could create meaningful risk if used badly, or where Anthropic believes unrestricted access could accelerate misuse or capability proliferation. Anthropic says these safeguards trigger in less than 5% of sessions on average, meaning most users should experience Fable as the full Mythos-class model most of the time. But that 5% is where the entire debate lives. Because if you are building a normal app, analyzing documents, writing code, doing finance work, or working on general research, Fable 5 may simply feel like a stronger frontier model. If you are doing security research, advanced biology, chemistry, or frontier model development, the product experience becomes more complicated. That is where Mythos 5 comes in. Mythos 5 is the same underlying model as Fable 5, but with safeguards lifted in some areas. It is not generally available. It is being deployed through Project Glasswing, Anthropic’s initiative with cyber defenders and critical software infrastructure providers. Anthropic says it plans to expand access through a broader trusted program. This is a meaningful shift from “everyone gets the same model” to “capability access depends on trust, use case, and risk category.” I do not think this is just product packaging. It is probably a preview of how frontier AI gets distributed from here. In crypto, we are used to open access as a cultural default. The whole industry is built around permissionless infrastructure, public networks, open liquidity, composability, and adversarial testing. The assumption is that if something is powerful, the network should expose it, and the market should figure out what survives. Frontier AI is moving in a different direction. The most capable systems are becoming too useful to keep entirely closed, but too risky to release without restrictions. That creates a middle layer: broad public access for most tasks, gated access for sensitive domains, and institutional partnerships for the highest-risk capabilities. There is a strong argument for this. If a model is genuinely good at finding and exploiting software vulnerabilities, then unrestricted release has obvious downside. Anthropic previously said Mythos Preview had found thousands of high-severity vulnerabilities, including some in major operating systems and browsers. Even if we treat that as an Anthropic claim rather than independent proof, the direction is clear: models are becoming serious cyber tools. That means the same capability can be defensive or offensive depending on who holds it. A security team using Mythos to audit critical infrastructure is very different from an unknown actor using it to automate exploit discovery. A biology researcher using the model to generate therapeutic hypotheses is very different from someone trying to gain dangerous biological uplift. The difficult part is that the boundary is not clean. Dual-use work is messy. Real security research can look like offensive security. Real biology can overlap with sensitive methods. Real AI research can look like distillation or capability extraction. If the classifier is too narrow, malicious users get through. If it is too broad, legitimate researchers get blocked or silently downgraded. This is why the transparency issue matters. After launch, Anthropic already faced backlash around invisible safeguards for frontier LLM development. The criticism was not only that the model had restrictions. Most serious users understand that frontier systems will have restrictions. The criticism was that some interventions were not visible enough to the user, which makes evaluation harder and damages trust. If a model refuses, that is annoying but clear. If a model falls back to a weaker model and tells you, that is also clear. But if a model quietly changes behavior, limits effectiveness, or routes around your task without making the intervention obvious, then developers cannot properly evaluate it. Researchers cannot know whether they are testing model capability, product policy, or invisible steering. That is a major problem. To Anthropic’s credit, they appear to have recognized this quickly and said they are changing Fable 5’s safeguards for frontier LLM development to make them visible. That is the right direction. Still, the tension does not disappear. The bigger question is whether frontier AI companies can build trust while also reserving the most powerful capabilities for trusted actors. This is not just about Anthropic. It is about the governance model of the entire AI stack. The public wants access. Developers want predictable behavior. Researchers want measurable capability. Governments want security. Labs want to avoid catastrophic misuse. Competitors want fair evaluation. Enterprises want privacy, reliability, and compliance. All of those demands collide inside a release like Fable / Mythos. Another under-discussed piece is data retention. Anthropic says Mythos-class traffic requires 30-day retention for safety monitoring, while also saying the data will not be used to train new Claude models and will be deleted after 30 days in almost all cases. That may be reasonable from a safety perspective, especially if the goal is detecting jailbreaks or coordinated misuse across many requests. But for enterprises, regulated industries, and sensitive research teams, it becomes a real deployment consideration. The more capable the model, the more likely users want to use it on sensitive work. The more sensitive the work, the more important retention policy becomes. So the model is not just competing on intelligence anymore. It is competing on governance. This is probably where the AI market is going. The best model will not simply be the one with the highest benchmark score. It will be the one that offers the best combination of capability, transparency, access control, reliability, compliance, cost, and trust. Fable 5 and Mythos 5 are interesting because they expose that full stack at once. There is the capability story: a model above Opus-class, built for long-horizon tasks and advanced reasoning. There is the safety story: classifiers, fallbacks, red-teaming, limited access, and trusted programs. There is the product story: public users get Fable, vetted users get Mythos. There is the trust story: users need to know when they are interacting with full capability and when safeguards are shaping the output. There is the market story: frontier AI is becoming less like a normal SaaS product and more like critical infrastructure. Personally, I think this release is one of the clearest signs that “open vs closed” is no longer the only useful framing. The new framing is closer to: who gets which capability, under what conditions, with what monitoring, and with what disclosure? That is less clean than the old debate, but probably more accurate. Based on the available information, Fable 5 may become an important public frontier model. Mythos 5 may become an important restricted capability layer for security and science. But the bigger experiment is the access model itself. If Anthropic gets the balance right, this could become a template for deploying very powerful AI safely while still letting most users benefit from the capability. If they get it wrong, it becomes a trust problem: too much opacity for developers, too much restriction for researchers, and too much central control over frontier capability. Either way, this is worth watching. Not just because Mythos looks powerful. Because it shows how AI labs may decide who is allowed to use power at all.

"Most of the online actions that we take on the internet will be run by ZK." In a recent episode, we chatted with @reka_eth from @RiscZero about the future of ZK within our online world. Find the full episode at the links below!